Mirrors
/
EveryDream2trainer
mirror of https://github.com/victorchall/EveryDream2trainer.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Update BASEMODELS.md 

upcast attn for sd2 conv, not critical but technically correct
This commit is contained in:
Victor Hall 2023-02-14 12:41:50 -05:00 committed by GitHub
parent 43f7f3c0f1
commit c0e47e5eab
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 2 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
doc/BASEMODELS.md
View File

@ -52,7 +52,7 @@ And the SD2.1 768 model (uses v2-v yaml and "v_prediction" prediction type):
--image_size 768 ^
--checkpoint_path v2-1_768-nonema-pruned.ckpt ^
--prediction_type v_prediction ^
--upcast_attn False ^
--upcast_attn True ^
--dump_path "ckpt_cache/v2-1_768-nonema-pruned"
And finally the SD2.0 512 base model (generally not recommended base model):
@ -62,7 +62,7 @@ And finally the SD2.0 512 base model (generally not recommended base model):
--image_size 512 ^
--checkpoint_path 512-base-ema.ckpt ^
--prediction_type epsilon ^
--upcast_attn False ^
--upcast_attn True ^
--dump_path "ckpt_cache/512-base-ema"
If you have other models, you need to know the base model that was used for them, **in particular use the correct yaml (original_config_file) or it will not properly convert.** Make sure to put some sort of name in the dump_path after "ckpt_cache/" so you can reference it later.